In [9]:
#visualizing the missing values
msno.bar(cr_df, color='green', sort='ascending', figsize=(15,5), fontsize=15)
Out[9]:
<AxesSubplot:>
Project type: Classification
Contributor: Shubham Harde
A Card issuing Bank has over issued its cash and credit card in-order-to increase its market share, even to unqualified applicants. At the same time, most cardholders, irrespective of their repayment ability, overused credit card for consumption and accumulated heavy credit and cash– card debts. The crisis caused the blow to consumer finance confidence and it is a big challenge for both banks and cardholders. From the perspective of risk control, estimating the probability of default will be more meaningful than classifying customers into the binary results – risky and non-risky.
This research employed a binary variable – default payment, Y (Yes = 1, No = 0), as the response variable.
This study used the following 23 variables as explanatory variables:
X1: Amount of the given credit (NT dollar): it includes both the individual consumer credit and his/her family (supplementary) credit
X2: Gender (1 = male; 2 = female)
X3: Education (1 = graduate school; 2 = university; 3 = high school; 4 = others)
X4: Marital status (1 = married; 2 = single; 3 = others)
X5: Age (year).
X6–X11: History of past payment. We tracked the past monthly payment records (from April to September, 2005) as follows:
X6 = the repayment status in September, 2005
X7 = the repayment status in August, 2005
X11 = the repayment status in April, 2005
The measurement scale for the repayment status is:
0 = pay duly;
1 = payment delay for one month
2 = payment delay for two months
8 = payment delay for eight months
9 = payment delay for nine months and above
X12–X17: Amount of bill statement (NT dollar)
X12 = amount of bill statement in September,2005
X13 = amount of bill statement in August, 2005
X17 = amount of bill statement in April, 2005
X18–X23: Amount of previous payment (NT dollar)
X18 = amount paid in September,2005
X19 = amount paid in August,2005
X23 = amount paid in April, 2005
• Analyze the trend on outstanding amount for the bank
• Is there any relationship between in outstanding amount / trend with respect to age, education, marriage, credit limit
• Does outstanding amount / trend affect the default behavior in next month.
• Identify the errors in dataset
• Clean those errors in the dataset.
• Try to Apply Statistics like EDA, Confidence Interval, Probability Distribution & Hypothesis
• Build ML Model using Logistic Regression & Decision Tree Classifier and Apply Feature Selection Techniques, Feature Scaling & Label Encoding Techniques
#Import Libraries
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import plotly.graph_objects as go
import plotly.express as px
import seaborn as sns
import missingno as msno
%matplotlib inline
#Library of warnings would assist to ignore warning issued
import warnings
warnings.filterwarnings('ignore')
#Import necessary statistical libraries
import scipy.stats as stats
import statsmodels.api as sm
from scipy.stats import norm
from scipy.stats import chisquare
from scipy.stats import chi2_contingency
from scipy.stats import ks_2samp
# import necessary libraries for feature engineering
from sklearn.model_selection import train_test_split
from imblearn.over_sampling import SMOTE
from sklearn.preprocessing import StandardScaler
#Import libraries for machine learning model
from sklearn.linear_model import LogisticRegression
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import RandomForestClassifier,AdaBoostClassifier
from xgboost import XGBClassifier
#load dataset
cr_df = pd.read_excel("F:/DS2024/Projects/default_of_credit_card_clients_0.xlsx")
Dataset First View
cr_df.head()
| ID | LIMIT_BAL | SEX | EDUCATION | MARRIAGE | AGE | PAY_0 | PAY_2 | PAY_3 | PAY_4 | ... | BILL_AMT4 | BILL_AMT5 | BILL_AMT6 | PAY_AMT1 | PAY_AMT2 | PAY_AMT3 | PAY_AMT4 | PAY_AMT5 | PAY_AMT6 | default payment next month | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | 20000 | 2 | 2 | 1 | 24 | 2 | 2 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 689 | 0 | 0 | 0 | 0 | 1 |
| 1 | 2 | 120000 | 2 | 2 | 2 | 26 | 0 | 2 | 0 | 0 | ... | 3272 | 3455 | 3261 | 0 | 1000 | 1000 | 1000 | 0 | 2000 | 1 |
| 2 | 3 | 90000 | 2 | 2 | 2 | 34 | 0 | 0 | 0 | 0 | ... | 14331 | 14948 | 15549 | 1518 | 1500 | 1000 | 1000 | 1000 | 5000 | 0 |
| 3 | 4 | 50000 | 2 | 2 | 1 | 37 | 0 | 0 | 0 | 0 | ... | 28314 | 28959 | 29547 | 2000 | 2019 | 1200 | 1100 | 1069 | 1000 | 0 |
| 4 | 5 | 50000 | 1 | 2 | 1 | 57 | 0 | 0 | 0 | 0 | ... | 20940 | 19146 | 19131 | 2000 | 36681 | 10000 | 9000 | 689 | 679 | 0 |
5 rows × 25 columns
Dataset Rows & Column Count ▼
#dataset rows & column count
cr_df.shape
(30000, 25)
the dataset has 30000 rows and 25 columns
Dataset Information ▼
cr_df.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 30000 entries, 0 to 29999 Data columns (total 25 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 ID 30000 non-null int64 1 LIMIT_BAL 30000 non-null int64 2 SEX 30000 non-null int64 3 EDUCATION 30000 non-null int64 4 MARRIAGE 30000 non-null int64 5 AGE 30000 non-null int64 6 PAY_0 30000 non-null int64 7 PAY_2 30000 non-null int64 8 PAY_3 30000 non-null int64 9 PAY_4 30000 non-null int64 10 PAY_5 30000 non-null int64 11 PAY_6 30000 non-null int64 12 BILL_AMT1 30000 non-null int64 13 BILL_AMT2 30000 non-null int64 14 BILL_AMT3 30000 non-null int64 15 BILL_AMT4 30000 non-null int64 16 BILL_AMT5 30000 non-null int64 17 BILL_AMT6 30000 non-null int64 18 PAY_AMT1 30000 non-null int64 19 PAY_AMT2 30000 non-null int64 20 PAY_AMT3 30000 non-null int64 21 PAY_AMT4 30000 non-null int64 22 PAY_AMT5 30000 non-null int64 23 PAY_AMT6 30000 non-null int64 24 default payment next month 30000 non-null int64 dtypes: int64(25) memory usage: 5.7 MB
cr_df.describe(include='all').T
| count | mean | std | min | 25% | 50% | 75% | max | |
|---|---|---|---|---|---|---|---|---|
| ID | 30000.0 | 15000.500000 | 8660.398374 | 1.0 | 7500.75 | 15000.5 | 22500.25 | 30000.0 |
| LIMIT_BAL | 30000.0 | 167484.322667 | 129747.661567 | 10000.0 | 50000.00 | 140000.0 | 240000.00 | 1000000.0 |
| SEX | 30000.0 | 1.603733 | 0.489129 | 1.0 | 1.00 | 2.0 | 2.00 | 2.0 |
| EDUCATION | 30000.0 | 1.853133 | 0.790349 | 0.0 | 1.00 | 2.0 | 2.00 | 6.0 |
| MARRIAGE | 30000.0 | 1.551867 | 0.521970 | 0.0 | 1.00 | 2.0 | 2.00 | 3.0 |
| AGE | 30000.0 | 35.485500 | 9.217904 | 21.0 | 28.00 | 34.0 | 41.00 | 79.0 |
| PAY_0 | 30000.0 | 0.356767 | 0.760594 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_2 | 30000.0 | 0.320033 | 0.801727 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_3 | 30000.0 | 0.304067 | 0.790589 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_4 | 30000.0 | 0.258767 | 0.761113 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_5 | 30000.0 | 0.221500 | 0.717720 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_6 | 30000.0 | 0.226567 | 0.715438 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| BILL_AMT1 | 30000.0 | 51223.330900 | 73635.860576 | -165580.0 | 3558.75 | 22381.5 | 67091.00 | 964511.0 |
| BILL_AMT2 | 30000.0 | 49179.075167 | 71173.768783 | -69777.0 | 2984.75 | 21200.0 | 64006.25 | 983931.0 |
| BILL_AMT3 | 30000.0 | 47013.154800 | 69349.387427 | -157264.0 | 2666.25 | 20088.5 | 60164.75 | 1664089.0 |
| BILL_AMT4 | 30000.0 | 43262.948967 | 64332.856134 | -170000.0 | 2326.75 | 19052.0 | 54506.00 | 891586.0 |
| BILL_AMT5 | 30000.0 | 40311.400967 | 60797.155770 | -81334.0 | 1763.00 | 18104.5 | 50190.50 | 927171.0 |
| BILL_AMT6 | 30000.0 | 38871.760400 | 59554.107537 | -339603.0 | 1256.00 | 17071.0 | 49198.25 | 961664.0 |
| PAY_AMT1 | 30000.0 | 5663.580500 | 16563.280354 | 0.0 | 1000.00 | 2100.0 | 5006.00 | 873552.0 |
| PAY_AMT2 | 30000.0 | 5921.163500 | 23040.870402 | 0.0 | 833.00 | 2009.0 | 5000.00 | 1684259.0 |
| PAY_AMT3 | 30000.0 | 5225.681500 | 17606.961470 | 0.0 | 390.00 | 1800.0 | 4505.00 | 896040.0 |
| PAY_AMT4 | 30000.0 | 4826.076867 | 15666.159744 | 0.0 | 296.00 | 1500.0 | 4013.25 | 621000.0 |
| PAY_AMT5 | 30000.0 | 4799.387633 | 15278.305679 | 0.0 | 252.50 | 1500.0 | 4031.50 | 426529.0 |
| PAY_AMT6 | 30000.0 | 5215.502567 | 17777.465775 | 0.0 | 117.75 | 1500.0 | 4000.00 | 528666.0 |
| default payment next month | 30000.0 | 0.221200 | 0.415062 | 0.0 | 0.00 | 0.0 | 0.00 | 1.0 |
Duplicates Values ▼
#dataset dupliucate value count
cr_df.duplicated().value_counts()
False 30000 dtype: int64
Missing Values/Null Values ▼
#missing values and null values count
null_values = cr_df.isnull().sum()
null_values
ID 0 LIMIT_BAL 0 SEX 0 EDUCATION 0 MARRIAGE 0 AGE 0 PAY_0 0 PAY_2 0 PAY_3 0 PAY_4 0 PAY_5 0 PAY_6 0 BILL_AMT1 0 BILL_AMT2 0 BILL_AMT3 0 BILL_AMT4 0 BILL_AMT5 0 BILL_AMT6 0 PAY_AMT1 0 PAY_AMT2 0 PAY_AMT3 0 PAY_AMT4 0 PAY_AMT5 0 PAY_AMT6 0 default payment next month 0 dtype: int64
#visualizing the missing values
msno.bar(cr_df, color='green', sort='ascending', figsize=(15,5), fontsize=15)
<AxesSubplot:>
msno.matrix(cr_df)
<AxesSubplot:>
If missing values present then above graph will show white horizontal space with respect to the perticular bar
What did you know about your dataset?
#Dataset Columns
cr_df.columns
Index(['ID', 'LIMIT_BAL', 'SEX', 'EDUCATION', 'MARRIAGE', 'AGE', 'PAY_0',
'PAY_2', 'PAY_3', 'PAY_4', 'PAY_5', 'PAY_6', 'BILL_AMT1', 'BILL_AMT2',
'BILL_AMT3', 'BILL_AMT4', 'BILL_AMT5', 'BILL_AMT6', 'PAY_AMT1',
'PAY_AMT2', 'PAY_AMT3', 'PAY_AMT4', 'PAY_AMT5', 'PAY_AMT6',
'default payment next month'],
dtype='object')
#Dataset Describe
cr_df.describe().T
| count | mean | std | min | 25% | 50% | 75% | max | |
|---|---|---|---|---|---|---|---|---|
| ID | 30000.0 | 15000.500000 | 8660.398374 | 1.0 | 7500.75 | 15000.5 | 22500.25 | 30000.0 |
| LIMIT_BAL | 30000.0 | 167484.322667 | 129747.661567 | 10000.0 | 50000.00 | 140000.0 | 240000.00 | 1000000.0 |
| SEX | 30000.0 | 1.603733 | 0.489129 | 1.0 | 1.00 | 2.0 | 2.00 | 2.0 |
| EDUCATION | 30000.0 | 1.853133 | 0.790349 | 0.0 | 1.00 | 2.0 | 2.00 | 6.0 |
| MARRIAGE | 30000.0 | 1.551867 | 0.521970 | 0.0 | 1.00 | 2.0 | 2.00 | 3.0 |
| AGE | 30000.0 | 35.485500 | 9.217904 | 21.0 | 28.00 | 34.0 | 41.00 | 79.0 |
| PAY_0 | 30000.0 | 0.356767 | 0.760594 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_2 | 30000.0 | 0.320033 | 0.801727 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_3 | 30000.0 | 0.304067 | 0.790589 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_4 | 30000.0 | 0.258767 | 0.761113 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_5 | 30000.0 | 0.221500 | 0.717720 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| PAY_6 | 30000.0 | 0.226567 | 0.715438 | 0.0 | 0.00 | 0.0 | 0.00 | 8.0 |
| BILL_AMT1 | 30000.0 | 51223.330900 | 73635.860576 | -165580.0 | 3558.75 | 22381.5 | 67091.00 | 964511.0 |
| BILL_AMT2 | 30000.0 | 49179.075167 | 71173.768783 | -69777.0 | 2984.75 | 21200.0 | 64006.25 | 983931.0 |
| BILL_AMT3 | 30000.0 | 47013.154800 | 69349.387427 | -157264.0 | 2666.25 | 20088.5 | 60164.75 | 1664089.0 |
| BILL_AMT4 | 30000.0 | 43262.948967 | 64332.856134 | -170000.0 | 2326.75 | 19052.0 | 54506.00 | 891586.0 |
| BILL_AMT5 | 30000.0 | 40311.400967 | 60797.155770 | -81334.0 | 1763.00 | 18104.5 | 50190.50 | 927171.0 |
| BILL_AMT6 | 30000.0 | 38871.760400 | 59554.107537 | -339603.0 | 1256.00 | 17071.0 | 49198.25 | 961664.0 |
| PAY_AMT1 | 30000.0 | 5663.580500 | 16563.280354 | 0.0 | 1000.00 | 2100.0 | 5006.00 | 873552.0 |
| PAY_AMT2 | 30000.0 | 5921.163500 | 23040.870402 | 0.0 | 833.00 | 2009.0 | 5000.00 | 1684259.0 |
| PAY_AMT3 | 30000.0 | 5225.681500 | 17606.961470 | 0.0 | 390.00 | 1800.0 | 4505.00 | 896040.0 |
| PAY_AMT4 | 30000.0 | 4826.076867 | 15666.159744 | 0.0 | 296.00 | 1500.0 | 4013.25 | 621000.0 |
| PAY_AMT5 | 30000.0 | 4799.387633 | 15278.305679 | 0.0 | 252.50 | 1500.0 | 4031.50 | 426529.0 |
| PAY_AMT6 | 30000.0 | 5215.502567 | 17777.465775 | 0.0 | 117.75 | 1500.0 | 4000.00 | 528666.0 |
| default payment next month | 30000.0 | 0.221200 | 0.415062 | 0.0 | 0.00 | 0.0 | 0.00 | 1.0 |
This research employed a binary variable – default payment, Y (Yes = 1, No = 0), as the response variable. This study used the following 23 variables as explanatory variables:
ID__: Unique ID of each client
Limit_Bal: Amount of the given credit (NT dollar): it includes both the individual consumer credit and his/her family (supplementary) credit.
Gender: 1 = male; 2 = female
Education: 1 = graduate school; 2 = university; 3 = high school; 4 = others
Marital status: 1 = married; 2 = single; 3 = others
Age: year
History of past payment.
We tracked the past monthly payment records (from April to September, 2005) as follows:
PAY_0 = the repayment status in September, 2005
PAY_2 = the repayment status in August, 2005
PAY_3 = the repayment status in July, 2005
PAY_4 = the repayment status in June, 2005
PAY_5 = the repayment status in May, 2005
PAY_6 = the repayment status in April, 2005
The measurement scale for the repayment status is:
0 = pay duly; 1 = payment delay for one month; 2 = payment delay for two months; ...; 8 = payment delay for eight months; 9 = payment delay for nine months and above.
Amount of bill statement (NT dollar)
BILL_AMT1: amount of bill statement in September,2005
BILL_AMT2 = amount of bill statement in August, 2005
BILL_AMT3 = amount of bill statement in July, 2005
BILL_AMT4: amount of bill statement in June,2005
BILL_AMT5 = amount of bill statement in May, 2005
BILL_AMT6 = amount of bill statement in April, 2005
Amount of previous payment (NT dollar)
PAY_AMT1 = amount paid in September,2005;
PAY_AMT2 = amount paid in August,2005;
PAY_AMT3 = amount paid in July, 2005.
PAY_AMT4 = amount paid in June,2005;
PAY_AMT5 = amount paid in May,2005;
PAY_AMT6 = amount paid in April, 2005.
default payment next month: Default Payment (1=Yes, 0=No)
#check unique values for each variable from dataset
for i in cr_df.columns.to_list():
print("The unique values number count", i,"is", cr_df[i].nunique())
The unique values number count ID is 30000 The unique values number count LIMIT_BAL is 81 The unique values number count SEX is 2 The unique values number count EDUCATION is 7 The unique values number count MARRIAGE is 4 The unique values number count AGE is 56 The unique values number count PAY_0 is 9 The unique values number count PAY_2 is 9 The unique values number count PAY_3 is 9 The unique values number count PAY_4 is 9 The unique values number count PAY_5 is 8 The unique values number count PAY_6 is 8 The unique values number count BILL_AMT1 is 22723 The unique values number count BILL_AMT2 is 22346 The unique values number count BILL_AMT3 is 22026 The unique values number count BILL_AMT4 is 21548 The unique values number count BILL_AMT5 is 21010 The unique values number count BILL_AMT6 is 20604 The unique values number count PAY_AMT1 is 7943 The unique values number count PAY_AMT2 is 7899 The unique values number count PAY_AMT3 is 7518 The unique values number count PAY_AMT4 is 6937 The unique values number count PAY_AMT5 is 6897 The unique values number count PAY_AMT6 is 6939 The unique values number count default payment next month is 2
#creating a copy of dataset for analysyis purpose (EDA)
credit_df = cr_df.copy()
credit_df.columns
Index(['ID', 'LIMIT_BAL', 'SEX', 'EDUCATION', 'MARRIAGE', 'AGE', 'PAY_0',
'PAY_2', 'PAY_3', 'PAY_4', 'PAY_5', 'PAY_6', 'BILL_AMT1', 'BILL_AMT2',
'BILL_AMT3', 'BILL_AMT4', 'BILL_AMT5', 'BILL_AMT6', 'PAY_AMT1',
'PAY_AMT2', 'PAY_AMT3', 'PAY_AMT4', 'PAY_AMT5', 'PAY_AMT6',
'default payment next month'],
dtype='object')
#renaming the columns to appropriate names.
#renaming dependent variables
credit_df.rename(columns={'default payment next month':'IsDefaulter'}, inplace=True)
#changing name of some columns for simplicity and for better understanding
credit_df.rename(columns={'PAY_0':'REPAY_STS_SEPT', 'PAY_2':'REPAY_STS_AUG', 'PAY_3':'REPAY_STS_JUL', 'PAY_4':'REPAY_STS_JUN', 'PAY_5':'REPAY_STS_MAY', 'PAY_6':'REPAY_STS_APR'}, inplace=True)
credit_df.rename(columns={'BILL_AMT1':'BILL_AMT_SEPT', 'BILL_AMT2':'BILL_AMT_AUG', 'BILL_AMT3':'BILL_AMT_JUL', 'BILL_AMT4':'BILL_AMT_JUN', 'BILL_AMT5':'BILL_AMT_MAY', 'BILL_AMT6':'BILL_AMT_APR'}, inplace=True)
credit_df.rename(columns={'PAY_AMT1':'PAY_AMT_SEPT', 'PAY_AMT2':'PAY_AMT_AUG', 'PAY_AMT3':'PAY_AMT_JUL', 'PAY_AMT4':'PAY_AMT_JUN', 'PAY_AMT5':'PAY_AMT_MAY', 'PAY_AMT6':'PAY_AMT_APR'}, inplace=True)
#dropping id from dataset
credit_df.drop('ID', axis=1, inplace=True)
#replacing values with there labels
credit_df.replace({'SEX': {1:'Male', 2:'Female'}}, inplace=True)
credit_df.replace({'EDUCATION': {1:'Graduate School', 2:'University', 3:'High School', 4:'Others', 5:'Others', 6:'Others', 0:'Others'}}, inplace=True)
credit_df.replace({'MARRIAGE': {1:'Married', 2:'Single', 3:'Others', 0:'Others'}}, inplace=True)
credit_df.replace({"IsDefaulter": {1:'Yes', 0:'No'}}, inplace=True)
#data frame for EDA analysis
credit_df.head()
| LIMIT_BAL | SEX | EDUCATION | MARRIAGE | AGE | REPAY_STS_SEPT | REPAY_STS_AUG | REPAY_STS_JUL | REPAY_STS_JUN | REPAY_STS_MAY | ... | BILL_AMT_JUN | BILL_AMT_MAY | BILL_AMT_APR | PAY_AMT_SEPT | PAY_AMT_AUG | PAY_AMT_JUL | PAY_AMT_JUN | PAY_AMT_MAY | PAY_AMT_APR | IsDefaulter | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 20000 | Female | University | Married | 24 | 2 | 2 | 0 | 0 | 0 | ... | 0 | 0 | 0 | 0 | 689 | 0 | 0 | 0 | 0 | Yes |
| 1 | 120000 | Female | University | Single | 26 | 0 | 2 | 0 | 0 | 0 | ... | 3272 | 3455 | 3261 | 0 | 1000 | 1000 | 1000 | 0 | 2000 | Yes |
| 2 | 90000 | Female | University | Single | 34 | 0 | 0 | 0 | 0 | 0 | ... | 14331 | 14948 | 15549 | 1518 | 1500 | 1000 | 1000 | 1000 | 5000 | No |
| 3 | 50000 | Female | University | Married | 37 | 0 | 0 | 0 | 0 | 0 | ... | 28314 | 28959 | 29547 | 2000 | 2019 | 1200 | 1100 | 1069 | 1000 | No |
| 4 | 50000 | Male | University | Married | 57 | 0 | 0 | 0 | 0 | 0 | ... | 20940 | 19146 | 19131 | 2000 | 36681 | 10000 | 9000 | 689 | 679 | No |
5 rows × 24 columns
What Manipulation done and insight you found?
converting numerical values to categorical values for easy understanding
pip install pandas_profiling
Requirement already satisfied: pandas_profiling in c:\users\shubham harde\anaconda3\lib\site-packages (3.6.6) Requirement already satisfied: ydata-profiling in c:\users\shubham harde\anaconda3\lib\site-packages (from pandas_profiling) (4.1.2) Requirement already satisfied: phik<0.13,>=0.11.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (0.12.3) Requirement already satisfied: visions[type_image_path]==0.7.5 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (0.7.5) Requirement already satisfied: seaborn<0.13,>=0.10.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (0.11.2) Requirement already satisfied: pandas!=1.4.0,<1.6,>1.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (1.4.4) Requirement already satisfied: imagehash==4.3.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (4.3.1) Requirement already satisfied: matplotlib<3.7,>=3.2 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (3.5.2) Requirement already satisfied: numpy<1.24,>=1.16.0 in c:\users\shubham harde\appdata\roaming\python\python39\site-packages (from ydata-profiling->pandas_profiling) (1.23.5) Requirement already satisfied: PyYAML<6.1,>=5.0.0 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (6.0.1) Requirement already satisfied: pydantic<1.11,>=1.8.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (1.10.7) Requirement already satisfied: typeguard<2.14,>=2.13.2 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (2.13.3) Requirement already satisfied: multimethod<1.10,>=1.4 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (1.9.1) Requirement already satisfied: tqdm<4.65,>=4.48.2 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (4.64.1) Requirement already satisfied: htmlmin==0.1.12 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (0.1.12) Requirement already satisfied: jinja2<3.2,>=2.11.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (2.11.3) Requirement already satisfied: scipy<1.10,>=1.4.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (1.9.1) Requirement already satisfied: requests<2.29,>=2.24.0 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (2.28.1) Requirement already satisfied: statsmodels<0.14,>=0.13.2 in c:\users\shubham harde\anaconda3\lib\site-packages (from ydata-profiling->pandas_profiling) (0.13.2) Requirement already satisfied: PyWavelets in c:\users\shubham harde\anaconda3\lib\site-packages (from imagehash==4.3.1->ydata-profiling->pandas_profiling) (1.3.0) Requirement already satisfied: pillow in c:\users\shubham harde\anaconda3\lib\site-packages (from imagehash==4.3.1->ydata-profiling->pandas_profiling) (9.2.0) Requirement already satisfied: attrs>=19.3.0 in c:\users\shubham harde\anaconda3\lib\site-packages (from visions[type_image_path]==0.7.5->ydata-profiling->pandas_profiling) (21.4.0) Requirement already satisfied: tangled-up-in-unicode>=0.0.4 in c:\users\shubham harde\anaconda3\lib\site-packages (from visions[type_image_path]==0.7.5->ydata-profiling->pandas_profiling) (0.2.0) Requirement already satisfied: networkx>=2.4 in c:\users\shubham harde\anaconda3\lib\site-packages (from visions[type_image_path]==0.7.5->ydata-profiling->pandas_profiling) (2.8.4) Requirement already satisfied: MarkupSafe>=0.23 in c:\users\shubham harde\anaconda3\lib\site-packages (from jinja2<3.2,>=2.11.1->ydata-profiling->pandas_profiling) (2.0.1) Requirement already satisfied: pyparsing>=2.2.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from matplotlib<3.7,>=3.2->ydata-profiling->pandas_profiling) (3.0.9) Requirement already satisfied: python-dateutil>=2.7 in c:\users\shubham harde\anaconda3\lib\site-packages (from matplotlib<3.7,>=3.2->ydata-profiling->pandas_profiling) (2.8.2) Requirement already satisfied: packaging>=20.0 in c:\users\shubham harde\anaconda3\lib\site-packages (from matplotlib<3.7,>=3.2->ydata-profiling->pandas_profiling) (21.3) Requirement already satisfied: fonttools>=4.22.0 in c:\users\shubham harde\anaconda3\lib\site-packages (from matplotlib<3.7,>=3.2->ydata-profiling->pandas_profiling) (4.25.0) Requirement already satisfied: cycler>=0.10 in c:\users\shubham harde\anaconda3\lib\site-packages (from matplotlib<3.7,>=3.2->ydata-profiling->pandas_profiling) (0.11.0) Requirement already satisfied: kiwisolver>=1.0.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from matplotlib<3.7,>=3.2->ydata-profiling->pandas_profiling) (1.4.2) Requirement already satisfied: pytz>=2020.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from pandas!=1.4.0,<1.6,>1.1->ydata-profiling->pandas_profiling) (2024.1) Requirement already satisfied: joblib>=0.14.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from phik<0.13,>=0.11.1->ydata-profiling->pandas_profiling) (1.3.2) Requirement already satisfied: typing-extensions>=4.2.0 in c:\users\shubham harde\anaconda3\lib\site-packages (from pydantic<1.11,>=1.8.1->ydata-profiling->pandas_profiling) (4.11.0) Requirement already satisfied: urllib3<1.27,>=1.21.1 in c:\users\shubham harde\anaconda3\lib\site-packages (from requests<2.29,>=2.24.0->ydata-profiling->pandas_profiling) (1.26.18) Requirement already satisfied: certifi>=2017.4.17 in c:\users\shubham harde\anaconda3\lib\site-packages (from requests<2.29,>=2.24.0->ydata-profiling->pandas_profiling) (2022.9.14) Requirement already satisfied: idna<4,>=2.5 in c:\users\shubham harde\anaconda3\lib\site-packages (from requests<2.29,>=2.24.0->ydata-profiling->pandas_profiling) (3.3) Requirement already satisfied: charset-normalizer<3,>=2 in c:\users\shubham harde\anaconda3\lib\site-packages (from requests<2.29,>=2.24.0->ydata-profiling->pandas_profiling) (2.0.4) Requirement already satisfied: patsy>=0.5.2 in c:\users\shubham harde\anaconda3\lib\site-packages (from statsmodels<0.14,>=0.13.2->ydata-profiling->pandas_profiling) (0.5.2) Requirement already satisfied: colorama in c:\users\shubham harde\anaconda3\lib\site-packages (from tqdm<4.65,>=4.48.2->ydata-profiling->pandas_profiling) (0.4.5) Requirement already satisfied: six in c:\users\shubham harde\anaconda3\lib\site-packages (from patsy>=0.5.2->statsmodels<0.14,>=0.13.2->ydata-profiling->pandas_profiling) (1.16.0) Note: you may need to restart the kernel to use updated packages.
WARNING: Ignoring invalid distribution -ygments (c:\users\shubham harde\anaconda3\lib\site-packages) WARNING: Ignoring invalid distribution -ygments (c:\users\shubham harde\anaconda3\lib\site-packages) WARNING: Ignoring invalid distribution -ygments (c:\users\shubham harde\anaconda3\lib\site-packages) WARNING: Ignoring invalid distribution -ygments (c:\users\shubham harde\anaconda3\lib\site-packages) WARNING: Ignoring invalid distribution -ygments (c:\users\shubham harde\anaconda3\lib\site-packages) WARNING: Ignoring invalid distribution -ygments (c:\users\shubham harde\anaconda3\lib\site-packages)
from pandas_profiling import ProfileReport
Profile = ProfileReport(credit_df, title="Credit Profile Report", explorative=True)
Profile.to_file('output.html')
Summarize dataset: 0%| | 0/5 [00:00<?, ?it/s]
Generate report structure: 0%| | 0/1 [00:00<?, ?it/s]
Render HTML: 0%| | 0/1 [00:00<?, ?it/s]
Export report to file: 0%| | 0/1 [00:00<?, ?it/s]
Chart1: Dependent Variable Analysis
Chart1 - Dependent Variable Analysis
counts_defaulter = credit_df['IsDefaulter'].value_counts()
# Create pie chart trace
fig = go.Figure(data=[go.Pie(labels=['Nondefaulter', 'Defaulter'], values=counts_defaulter.values)])
fig.update_layout(title_text="Pie Chart of Defaulter Status")
fig.show()
Pie Chart used to represent the summarizing set of nominal data or displays the different values of a given categorical vaariable(Percentage Distribution)
From the graph we can see that number of defaulter(22.1%) is less than the non defaulter(77.9%)
3.Gained insight from the pie chart helps to create positive business impact?
Yes, class imbalance can have -ve impact on classification model accurary, so we must address this issue before putting the data into training
Chart2 - Visualization of Gender Column
defaulter_counts = credit_df.groupby(['SEX', 'IsDefaulter']).size().reset_index(name='Count')
# Univariate analysis of gender column
fig1 = px.pie(values=credit_df['SEX'].value_counts(),names=['Female', 'Male'],title='Distribution Male and Female')
# Bivariate analysis Sex vs Defaulters
fig2 = px.bar(data_frame=defaulter_counts,x='SEX',y='Count',color='IsDefaulter',barmode='stack',
title='Bivariate Analysis: SEX VS IS DEFAULTERS',
labels={'SEX': 'SEX', 'Count': 'Count', 'IsDefaulter': 'Defaulters'})
fig1.show()
fig2.show()
TEST = credit_df.groupby(by=['SEX', 'IsDefaulter'])
count_values = TEST['IsDefaulter'].count()
count_values
SEX IsDefaulter
Female No 14349
Yes 3763
Male No 9015
Yes 2873
Name: IsDefaulter, dtype: int64
pie chart analysis percentage distribution of a feature and bar chart is to compare the feature column values with dependent variable
Female customers are more than the male customer // There are 3763 female customer defaulters and 2874 male customer defaulters
As the positive side female customers are better customers compared to male as the non defualters are maximum lies in female category // no negative impact
Chart3 - Education column Vs Defaulters
defaulter_counts = credit_df.groupby(['EDUCATION', 'IsDefaulter']).size().reset_index(name='Count')
# Univariate analysis of gender column
fig1 = px.pie(values=credit_df['EDUCATION'].value_counts(),names=['Graduate School','University','High School','Others'],title='Distribution of Education Qualification of customers')
# Bivariate analysis Sex vs Defaulters
fig2 = px.bar(data_frame=defaulter_counts,x='EDUCATION',y='Count',color='IsDefaulter',barmode='stack',
title='Bivariate Analysis: EDUCATION VS IS DEFAULTERS',
labels={'EDUCATION': 'EDUCATION', 'Count': 'Count', 'IsDefaulter': 'Defaulters'})
fig1.show()
fig2.show()
TEST = credit_df.groupby(by=['EDUCATION', 'IsDefaulter'])
count_values = TEST['IsDefaulter'].count()
count_values
EDUCATION IsDefaulter
Graduate School No 8549
Yes 2036
High School No 3680
Yes 1237
Others No 435
Yes 33
University No 10700
Yes 3330
Name: IsDefaulter, dtype: int64
Insight
Most of the customers educational qualification is university pass out followed by graduate school pass out // most numbers of defualters is also from university but in terms of default ratio university people having less default ratio, high school have most default ratio
Chart 4 - Visualization of Mariage column
#chart 4 - visualization code
defaulter_counts = credit_df.groupby(['MARRIAGE', 'IsDefaulter']).size().reset_index(name='Count')
# Univariate analysis of gender column
fig1 = px.pie(values=credit_df['MARRIAGE'].value_counts(),names=['Married', 'Single', 'Others'],title='Distribution Marriage Column')
# Bivariate analysis Sex vs Defaulters
fig2 = px.bar(data_frame=defaulter_counts,x='MARRIAGE',y='Count',color='IsDefaulter',barmode='stack',
title='Bivariate Analysis: MARRIAGE VS IS DEFAULTERS',
labels={'MARRIAGE': 'MARRIAGE', 'Count': 'Count', 'IsDefaulter': 'Defaulters'})
fig1.show()
fig2.show()
TEST = credit_df.groupby(by=['MARRIAGE', 'IsDefaulter'])
count = TEST['IsDefaulter'].value_counts()
count
MARRIAGE IsDefaulter IsDefaulter
Married No No 10453
Yes Yes 3206
Others No No 288
Yes Yes 89
Single No No 12623
Yes Yes 3341
Name: IsDefaulter, dtype: int64
Insight
Maximum card holder are single // Married people have higher default ratio
Business Impact
Singles spends more than married people thus they will use credit card more frequently. As a result, targeting single people will undoubtly enhance sales
Chart 5 - Visualization of age column
# Create a histogram for the distribution of Age
fig_age_dist = px.histogram(credit_df, x='AGE', title='Distribution of Age of Credit Card Holder')
# Update histogram traces to add borders
fig_age_dist.update_traces(marker_line_color='black', marker_line_width=1, opacity=0.7)
# Create two data frames of male and female
Age_male = credit_df[credit_df['SEX'] == 'Male']
Age_female = credit_df[credit_df['SEX'] == 'Female']
# Plot age of male card holder vs Defaulter with gap
fig_male = px.histogram(Age_male, x='AGE', color='IsDefaulter', title='Age of Male v/s Defaulters',
barmode='group', histnorm='percent',
category_orders={'IsDefaulter': ['No', 'Yes']})
# Plot age of female card holder vs Defaulter with gap
fig_female = px.histogram(Age_female, x='AGE', color='IsDefaulter', title='Age of Female v/s Defaulters',
barmode='group', histnorm='percent',
category_orders={'IsDefaulter': ['No', 'Yes']})
# Update layout to set axis labels and show all age numbers
fig_age_dist.update_layout(xaxis=dict(type='category', categoryorder='category ascending'),
xaxis_title='Age', yaxis_title='Count')
fig_male.update_layout(xaxis=dict(type='category', categoryorder='category ascending'),
xaxis_title='Age', yaxis_title='Count')
fig_female.update_layout(xaxis=dict(type='category', categoryorder='category ascending'),
xaxis_title='Age', yaxis_title='Count')
# Show the plots
fig_age_dist.show()
fig_male.show()
fig_female.show()
Insight
Most of the credit card holder belongs to 25-30 years
Number of credit card holder are less above 60 years
For equal are ratio of defaulter for male and female not varied much
Above 50 years old the ratio of defaulter had increased
Business Impact
Customers are fewer as the majority of defaulters are between the age of 20 and 25 and over the age of 60. Before issuing the credit card, it must be throughly reviewed. Otherwise, it may have a negative impact on carporate growth
Chart 6 - Visualization of Default Payment next month Column
# chart - 6 visualization code By using matplotlib and seaborn lib
fig, ax = plt.subplots(3, 1, figsize=(15, 25))
# univariate analysis - distribution of Marriage
age_dist = sns.histplot(x=credit_df['LIMIT_BAL'], ax=ax[0], bins=10)
age_dist.set_title('Distribution of Balance Limit of Credit Card Holder', fontsize=14, fontweight='bold')
Limit_1 = credit_df[credit_df['LIMIT_BAL'] <= 500000]
Limit_2 = credit_df[credit_df['LIMIT_BAL'] > 500000]
# plot age of male card holder vs Defaulter
sns.countplot(x='LIMIT_BAL', hue='IsDefaulter', data=Limit_1, ax=ax[1])
ax[1].set_title('Balance Limit v/s Defaulters', fontsize=14, fontweight='bold')
ax[1].tick_params(axis='x', labelrotation=60)
ax[1].set_xlabel('Balance Limit', fontsize=12)
ax[1].set_ylabel('Count', fontsize=12)
# plot age of male card holder vs Defaulter
sns.countplot(x='LIMIT_BAL', hue='IsDefaulter', data=Limit_2, ax=ax[2])
ax[2].set_title('Balance Limit v/s Defaulters', fontsize=14, fontweight='bold')
ax[2].tick_params(axis='x', labelrotation=60)
ax[2].set_xlabel('Balance Limit', fontsize=12)
ax[2].set_ylabel('Count', fontsize=12)
Text(0, 0.5, 'Count')
Insight:
Maximum credit card limit is below 100000
Very few credit limit are above 500000
When credit limit is less than equal to 50k default ration is high
default ratio is exceptionally high when credit limit is 550000 and 600000
Business Impact:
Customers with credit limit of $ 50,000 or more are more likely to default, so the bank should charge them in the event of a default and may cut their limit to avoid any form of damage to the company
Chart 7 - Monthly Repayment Status vs Defaulters
# Chart - 7 visualization code
Repay_col = ['REPAY_STS_SEPT' , 'REPAY_STS_AUG', 'REPAY_STS_JUL', 'REPAY_STS_JUN', 'REPAY_STS_MAY' , 'REPAY_STS_APR']
#plot the uniuvariate distrubtion
fig,ax=plt.subplots(len(Repay_col),1,figsize=(10,20))
for i , col in enumerate(Repay_col):
sns. countplot ( x=col, hue = 'IsDefaulter', data=credit_df, ax=ax[i])
ax[i].set_title(col)
plt.tight_layout()
plt.show()
for i in range(0, len(Repay_col)):
TEST = credit_df.groupby(by=[Repay_col[i], 'IsDefaulter'])
count_values = TEST['IsDefaulter'].count()
print(count_values)
REPAY_STS_SEPT IsDefaulter
0 No 19975
Yes 3207
1 No 2436
Yes 1252
2 No 823
Yes 1844
3 No 78
Yes 244
4 No 24
Yes 52
5 No 13
Yes 13
6 No 5
Yes 6
7 No 2
Yes 7
8 No 8
Yes 11
Name: IsDefaulter, dtype: int64
REPAY_STS_AUG IsDefaulter
0 No 21402
Yes 4160
1 No 23
Yes 5
2 No 1743
Yes 2184
3 No 125
Yes 201
4 No 49
Yes 50
5 No 10
Yes 15
6 No 3
Yes 9
7 No 8
Yes 12
8 No 1
Name: IsDefaulter, dtype: int64
REPAY_STS_JUL IsDefaulter
0 No 21353
Yes 4434
1 No 3
Yes 1
2 No 1850
Yes 1969
3 No 102
Yes 138
4 No 32
Yes 44
5 No 9
Yes 12
6 No 9
Yes 14
7 No 5
Yes 22
8 No 1
Yes 2
Name: IsDefaulter, dtype: int64
REPAY_STS_JUN IsDefaulter
0 No 21733
Yes 4757
1 No 1
Yes 1
2 No 1506
Yes 1653
3 No 70
Yes 110
4 No 23
Yes 46
5 No 17
Yes 18
6 No 3
Yes 2
7 No 10
Yes 48
8 No 1
Yes 1
Name: IsDefaulter, dtype: int64
REPAY_STS_MAY IsDefaulter
0 No 22045
Yes 4987
2 No 1203
Yes 1423
3 No 65
Yes 113
4 No 33
Yes 51
5 No 7
Yes 10
6 No 1
Yes 3
7 No 10
Yes 48
8 Yes 1
Name: IsDefaulter, dtype: int64
REPAY_STS_APR IsDefaulter
0 No 21896
Yes 5025
2 No 1365
Yes 1401
3 No 66
Yes 118
4 No 18
Yes 31
5 No 6
Yes 7
6 No 5
Yes 14
7 No 8
Yes 38
8 Yes 2
Name: IsDefaulter, dtype: int64
for col in Repay_col:
# Filter the DataFrame to include only rows where 'IsDefaulter' is 'yes '
filtered_df = credit_df[credit_df['IsDefaulter'] == 'Yes']
# Group the filtered DataFrame by the current column and count the 'IsDefaulter' values
TEST = filtered_df.groupby([col])['IsDefaulter'].count().reset_index(name='NO.of.Defaluters')
TEST = TEST.sort_values(by='NO.of.Defaluters', ascending=False)
print(TEST)
REPAY_STS_SEPT NO.of.Defaluters 0 0 3207 2 2 1844 1 1 1252 3 3 244 4 4 52 5 5 13 8 8 11 7 7 7 6 6 6 REPAY_STS_AUG NO.of.Defaluters 0 0 4160 2 2 2184 3 3 201 4 4 50 5 5 15 7 7 12 6 6 9 1 1 5 REPAY_STS_JUL NO.of.Defaluters 0 0 4434 2 2 1969 3 3 138 4 4 44 7 7 22 6 6 14 5 5 12 8 8 2 1 1 1 REPAY_STS_JUN NO.of.Defaluters 0 0 4757 2 2 1653 3 3 110 7 7 48 4 4 46 5 5 18 6 6 2 1 1 1 8 8 1 REPAY_STS_MAY NO.of.Defaluters 0 0 4987 1 2 1423 2 3 113 3 4 51 6 7 48 4 5 10 5 6 3 7 8 1 REPAY_STS_APR NO.of.Defaluters 0 0 5025 1 2 1401 2 3 118 6 7 38 3 4 31 5 6 14 4 5 7 7 8 2
Insight:
We can see that customers who pay only the minimum amount have a higher risk of default customers in this category also mostly use credit card
When payment delay is greater than 2 months default ratio is high
Chart 8 - Bill Amount Vs Defaulters
monthlybill_col = ['BILL_AMT_SEPT', 'BILL_AMT_AUG', 'BILL_AMT_JUL', 'BILL_AMT_JUN', 'BILL_AMT_MAY', 'BILL_AMT_APR']
# Chart - 8 visualization code
fig, ax = plt.subplots(len(monthlybill_col), 1, figsize=(10, 30))
for i, col in enumerate(monthlybill_col):
# plot the bivariate with isdefaulter
sns.boxplot(x=col, y='IsDefaulter', data=credit_df, ax=ax[i])
ax[i].set_title(col)
Why this chart?
Box plot provide a rapid visual assesment of a dataset's variability. They display the dataset's median, upper and lower qualities, lowest and maximum values, and any outliers in data show errors or uncommon events.
Insight:
from the plot we can see that distribution of monthly bill has no thick difference between defaulter and non defaulter
Business Impact:
There are no insight on businessfrom the figures
Chart 9 -Pay Amount vs Bill Amount
# Chart - 9 visualization code
previous_col = ['PAY_AMT_SEPT', 'PAY_AMT_AUG', 'PAY_AMT_JUL', 'PAY_AMT_JUN', 'PAY_AMT_MAY', 'PAY_AMT_APR']
# plot the univariate distribution
fig, ax = plt.subplots(len(previous_col), 1, figsize=(10, 30))
for i, col in enumerate(previous_col):
# plot the bivariate with isdefaulter
sns.scatterplot(x=col, y=monthlybill_col[i], hue='IsDefaulter', data=credit_df, ax=ax[i])
ax[i].set_title(col)
Insight:
When Previous payment is low but monthly bill is high there is high possibility of default the payment
Business Impact:
When customers is spending exceptionally high according to his payment history company should concerned
Chart 10 - Corelation Heat Map
#chart 10 - Visualization code
#Correlation Heatmap visualization code
plt.figure(figsize=(20, 10))
sns.heatmap(credit_df.corr(),linewidths=.5, annot=True, cmap='viridis')
plt.show()
Why this chart?
Correlation heatmaps are graphical representations of the strength of correlations between numerical data. Correlation plots are used to determine which variables are related to one another and how strong this relationship is. A correlation plot often has several numerical variables, each represented by a column. Each row represents the relationship between two variables. favourable numbers imply a favourable relationship, whereas negative values suggest a negative relationship.
Insight:
According to the heatmap, some features (for example, PAY_MAY & PAY_APR, BILL_AMT_MAY & BILL_AMT_APR, etc.) are highly associated to each other, but we are not going to eliminate any of them because they contain client transaction details.
Hypothetical Statement 1 : Married male defaults average age is 35 Years.
Null Hypothesis : N = 35
Alternate Hypothesis: N != 35
Test Type: Two Tailed Test
# Perform Statistical Test to obtain P-Value
# Perform Statistical Test to obtain P-Value
sample_1=credit_df[(credit_df['MARRIAGE' ]=='Married') & (credit_df['SEX' ]=='Male') & (credit_df['IsDefaulter' ]=='Yes')].sample(1000)
# Getting the required parameter values for hypothesis testing
N=35
sample_mean = sample_1['AGE'].mean()
size = len(sample_1)
# the standard deviation for population
std_pop = credit_df['AGE'].std()
Z_stat = ((sample_mean - N) / (std_pop / np.sqrt(N)))
# calculate the z value
z_value = norm.cdf(Z_stat, 0, 1)
# calculate p value
if z_value > 0.5:
p_value = 2 * (1 - z_value)
else:
p_value = 2 * z_value
print(f'P value is {p_value}')
if p_value >= 0.05:
print('Fail to reject the null hypothesis')
else:
print('Reject the null hypothesis')
P value is 9.953881684676702e-06 Reject the null hypothesis
mean_median_difference=sample_1["AGE"].mean()- sample_1["AGE"].median()
print("Mean Median Difference is : ",mean_median_difference)
Mean Median Difference is : 0.8840000000000003
Which statistical test have you done to obtain P-Value?
I utilized Z-test as a statistical test to produce P-value and discovered that the null hypothesis was rejected and that married male customers defaulting do not have an average age of 35 years
Hypothetical Statement 2
Null Hypothesis : mean = 90000
Alternate Hypothesis: mean != 90000
Test Type: Two Tailed Test
# Perform Statistical Test to obtain P-Value
# Perform Statistical Test to obtain P-Value
# Perform Statistical Test to obtain P-Value
sample_2=credit_df[(credit_df['IsDefaulter']=='Yes')].sample(1000)
# Getting the required parameter values for hypothesis testing
N =90000
sample_mean = sample_2['LIMIT_BAL'].mean()
size = len(sample_2)
# calculate the t_statistic and p_value
t_statistic,P_value=stats.ttest_1samp(sample_2['LIMIT_BAL'],N)
print('T-statistics value', t_statistic)
print("P-Value", P_value)
if P_value >=0.05:
print('Fail to reject the null hypothesis')
else:
print('Reject the null hypothesis')
T-statistics value 10.49656242764944 P-Value 1.6014985926171755e-24 Reject the null hypothesis
Which statistical test have you done to obtain P-value?
I utilized T-test as a statistical test to generate P-value and discovered that the null hypothesis was rejected the customers that defaulted had not an average credit limit 90000
mean_median_difference=sample_2['LIMIT_BAL'].mean()- sample_2['LIMIT_BAL'].median()
print("Mean Median Difference is : ",mean_median_difference)
Mean Median Difference is : 48040.0
The accompanying figure shows that the median is bigger than the mean over ten. As a result, the distribution is positively biassed. Z-Test cannot be conducted on skewed data.
Even with strongly skewed data, t t-tests and their related confidence intervals can and should be utilised in investigations with a large sample size.
So, with skewed data, we can utilise the T-test to get a better result. As a result, I used the t t - test.
1 State your research hypothesis as a null hypothesis and alternate hypothesis
Null Hypothesis(HO): Defaulter does not depends on Educational qualification
Alternate Hypothesis(Ha): Defaulter also get affected by educational qualification
2 Perform an appropriate statistical test >
# Perform Statistical Test to obtain P-Value
#Creating contigency table of EDUCATION AND DEFAULT PAYMENT NEXT MONTH
cont_table=pd.crosstab(credit_df['IsDefaulter'], credit_df['EDUCATION'], margins=False)
#Using Chi-square test to validate Null/Alternate Hypothesis
stat, P_value, dof, expected = chi2_contingency(cont_table)
print("P-value", P_value)
if P_value >=0.05:
print('Fail to reject the null hypothesis')
else:
print('Reject the null hypothesis')
P-value 1.495064564810615e-34 Reject the null hypothesis
Which statistical test have you done to obtain P-Value?
I used chi-square test to generate P_value and discoverd null hypothesis was rejected since defaulter does not depend on educational qualification.
Why did you choose the specific statistical test?
Here we have to test difference between defaulter and non defaulter with educational qualification.The Chi-Square test is a statistical procedure for determining the difference between observed and expected data.
#Handling Missing Values & Missing Values Imputation
Nullvalues = credit_df.isnull().sum()
Nullvalues
LIMIT_BAL 0 SEX 0 EDUCATION 0 MARRIAGE 0 AGE 0 REPAY_STS_SEPT 0 REPAY_STS_AUG 0 REPAY_STS_JUL 0 REPAY_STS_JUN 0 REPAY_STS_MAY 0 REPAY_STS_APR 0 BILL_AMT_SEPT 0 BILL_AMT_AUG 0 BILL_AMT_JUL 0 BILL_AMT_JUN 0 BILL_AMT_MAY 0 BILL_AMT_APR 0 PAY_AMT_SEPT 0 PAY_AMT_AUG 0 PAY_AMT_JUL 0 PAY_AMT_JUN 0 PAY_AMT_MAY 0 PAY_AMT_APR 0 IsDefaulter 0 dtype: int64
No Null Values present in data set
duplicatevalues = credit_df.duplicated()
duplicatevalues
0 False
1 False
2 False
3 False
4 False
...
29995 False
29996 False
29997 False
29998 False
29999 False
Length: 30000, dtype: bool
No duplicated values present in the data set
#Handling outliers & outlier treatments
credit_df.describe()
| LIMIT_BAL | AGE | REPAY_STS_SEPT | REPAY_STS_AUG | REPAY_STS_JUL | REPAY_STS_JUN | REPAY_STS_MAY | REPAY_STS_APR | BILL_AMT_SEPT | BILL_AMT_AUG | BILL_AMT_JUL | BILL_AMT_JUN | BILL_AMT_MAY | BILL_AMT_APR | PAY_AMT_SEPT | PAY_AMT_AUG | PAY_AMT_JUL | PAY_AMT_JUN | PAY_AMT_MAY | PAY_AMT_APR | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| count | 30000.000000 | 30000.000000 | 30000.000000 | 30000.000000 | 30000.000000 | 30000.000000 | 30000.00000 | 30000.000000 | 30000.000000 | 30000.000000 | 3.000000e+04 | 30000.000000 | 30000.000000 | 30000.000000 | 30000.000000 | 3.000000e+04 | 30000.00000 | 30000.000000 | 30000.000000 | 30000.000000 |
| mean | 167484.322667 | 35.485500 | 0.356767 | 0.320033 | 0.304067 | 0.258767 | 0.22150 | 0.226567 | 51223.330900 | 49179.075167 | 4.701315e+04 | 43262.948967 | 40311.400967 | 38871.760400 | 5663.580500 | 5.921163e+03 | 5225.68150 | 4826.076867 | 4799.387633 | 5215.502567 |
| std | 129747.661567 | 9.217904 | 0.760594 | 0.801727 | 0.790589 | 0.761113 | 0.71772 | 0.715438 | 73635.860576 | 71173.768783 | 6.934939e+04 | 64332.856134 | 60797.155770 | 59554.107537 | 16563.280354 | 2.304087e+04 | 17606.96147 | 15666.159744 | 15278.305679 | 17777.465775 |
| min | 10000.000000 | 21.000000 | 0.000000 | 0.000000 | 0.000000 | 0.000000 | 0.00000 | 0.000000 | -165580.000000 | -69777.000000 | -1.572640e+05 | -170000.000000 | -81334.000000 | -339603.000000 | 0.000000 | 0.000000e+00 | 0.00000 | 0.000000 | 0.000000 | 0.000000 |
| 25% | 50000.000000 | 28.000000 | 0.000000 | 0.000000 | 0.000000 | 0.000000 | 0.00000 | 0.000000 | 3558.750000 | 2984.750000 | 2.666250e+03 | 2326.750000 | 1763.000000 | 1256.000000 | 1000.000000 | 8.330000e+02 | 390.00000 | 296.000000 | 252.500000 | 117.750000 |
| 50% | 140000.000000 | 34.000000 | 0.000000 | 0.000000 | 0.000000 | 0.000000 | 0.00000 | 0.000000 | 22381.500000 | 21200.000000 | 2.008850e+04 | 19052.000000 | 18104.500000 | 17071.000000 | 2100.000000 | 2.009000e+03 | 1800.00000 | 1500.000000 | 1500.000000 | 1500.000000 |
| 75% | 240000.000000 | 41.000000 | 0.000000 | 0.000000 | 0.000000 | 0.000000 | 0.00000 | 0.000000 | 67091.000000 | 64006.250000 | 6.016475e+04 | 54506.000000 | 50190.500000 | 49198.250000 | 5006.000000 | 5.000000e+03 | 4505.00000 | 4013.250000 | 4031.500000 | 4000.000000 |
| max | 1000000.000000 | 79.000000 | 8.000000 | 8.000000 | 8.000000 | 8.000000 | 8.00000 | 8.000000 | 964511.000000 | 983931.000000 | 1.664089e+06 | 891586.000000 | 927171.000000 | 961664.000000 | 873552.000000 | 1.684259e+06 | 896040.00000 | 621000.000000 | 426529.000000 | 528666.000000 |
#outlier detection plot
for col in credit_df.describe().columns:
plt.figure(figsize=(12,8))
sns.boxplot(x=credit_df[col], color='r')
What all outlier treatment techniques have you used and why did you use those techniques?
I used the boxplot to look for outliers in the dataset. Some points appear to be outliers in numeric features that are primarily connected to quantity, and we can use the IQR method to treat these outliers without knowing the exact limit of these columns from the bank, and addressing these outlying amounts will result in data loss.And there is a potential that the amount that was maximal and appeared tdbe an outlier was the true value for that characteristic, thus I did not employ any approaches to treat these outlying numbers.
df = credit_df.copy() df.info() #From imfo we can see that there 3 object type nad reamining are int type we need to change theat type in to int by cateroigal encodig
df.replace({'IsDefaulter ' : {'Yes':1, 'No': 0}}, inplace=True)
df=pd.get_dummies (df) df. head()
df = credit_df.copy()
df.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 30000 entries, 0 to 29999 Data columns (total 24 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 LIMIT_BAL 30000 non-null int64 1 SEX 30000 non-null object 2 EDUCATION 30000 non-null object 3 MARRIAGE 30000 non-null object 4 AGE 30000 non-null int64 5 REPAY_STS_SEPT 30000 non-null int64 6 REPAY_STS_AUG 30000 non-null int64 7 REPAY_STS_JUL 30000 non-null int64 8 REPAY_STS_JUN 30000 non-null int64 9 REPAY_STS_MAY 30000 non-null int64 10 REPAY_STS_APR 30000 non-null int64 11 BILL_AMT_SEPT 30000 non-null int64 12 BILL_AMT_AUG 30000 non-null int64 13 BILL_AMT_JUL 30000 non-null int64 14 BILL_AMT_JUN 30000 non-null int64 15 BILL_AMT_MAY 30000 non-null int64 16 BILL_AMT_APR 30000 non-null int64 17 PAY_AMT_SEPT 30000 non-null int64 18 PAY_AMT_AUG 30000 non-null int64 19 PAY_AMT_JUL 30000 non-null int64 20 PAY_AMT_JUN 30000 non-null int64 21 PAY_AMT_MAY 30000 non-null int64 22 PAY_AMT_APR 30000 non-null int64 23 IsDefaulter 30000 non-null object dtypes: int64(20), object(4) memory usage: 5.5+ MB
df.columns
Index(['LIMIT_BAL', 'SEX', 'EDUCATION', 'MARRIAGE', 'AGE', 'REPAY_STS_SEPT',
'REPAY_STS_AUG', 'REPAY_STS_JUL', 'REPAY_STS_JUN', 'REPAY_STS_MAY',
'REPAY_STS_APR', 'BILL_AMT_SEPT', 'BILL_AMT_AUG', 'BILL_AMT_JUL',
'BILL_AMT_JUN', 'BILL_AMT_MAY', 'BILL_AMT_APR', 'PAY_AMT_SEPT',
'PAY_AMT_AUG', 'PAY_AMT_JUL', 'PAY_AMT_JUN', 'PAY_AMT_MAY',
'PAY_AMT_APR', 'IsDefaulter'],
dtype='object')
credit_df.columns
Index(['LIMIT_BAL', 'SEX', 'EDUCATION', 'MARRIAGE', 'AGE', 'REPAY_STS_SEPT',
'REPAY_STS_AUG', 'REPAY_STS_JUL', 'REPAY_STS_JUN', 'REPAY_STS_MAY',
'REPAY_STS_APR', 'BILL_AMT_SEPT', 'BILL_AMT_AUG', 'BILL_AMT_JUL',
'BILL_AMT_JUN', 'BILL_AMT_MAY', 'BILL_AMT_APR', 'PAY_AMT_SEPT',
'PAY_AMT_AUG', 'PAY_AMT_JUL', 'PAY_AMT_JUN', 'PAY_AMT_MAY',
'PAY_AMT_APR', 'IsDefaulter'],
dtype='object')
4.1 Feature Manipulation
#Manipulate Features to minimize feature correlation and create new features
4.2 Feature Selection
import pandas as pd
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import chi2
from sklearn.preprocessing import MinMaxScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
# Load your dataset (replace 'credit.csv' with the actual path to your dataset)
data = credit_df
# Assuming your target variable is 'IsDefaulter' and other columns are features
X = data.drop('IsDefaulter', axis=1)
y = data['IsDefaulter']
# Separate numerical and categorical columns
numeric_cols = X.select_dtypes(include=['number']).columns
categorical_cols = X.select_dtypes(include=['object']).columns
# Preprocessing: Scale numerical features and one-hot encode categorical features
preprocessor = ColumnTransformer(
transformers=[
('num', MinMaxScaler(), numeric_cols),
('cat', OneHotEncoder(), categorical_cols)
])
# Apply preprocessing to X
X_scaled = preprocessor.fit_transform(X)
# Initialize the SelectKBest object with chi2 as the scoring function
# You can specify the number of top features you want to select using the 'k' parameter
k_best = SelectKBest(score_func=chi2, k='all')
# Fit the SelectKBest object to your data
k_best.fit(X_scaled, y)
# Get the feature scores
feature_scores = k_best.scores_
# Create a DataFrame to display the feature names and their corresponding Chi-squared scores
all_feature_names = list(numeric_cols) + list(preprocessor.named_transformers_['cat'].get_feature_names_out(categorical_cols))
feature_chi2_scores = pd.DataFrame({'Feature': all_feature_names, 'Chi-squared_Score': feature_scores})
# Sort the features by chi-squared score in descending order
feature_chi2_scores = feature_chi2_scores.sort_values(by='Chi-squared_Score', ascending=False)
# Display the top features with their chi-squared scores
print(feature_chi2_scores)
Feature Chi-squared_Score 2 REPAY_STS_SEPT 953.610288 3 REPAY_STS_AUG 805.782247 4 REPAY_STS_JUL 634.908960 5 REPAY_STS_JUN 607.700938 6 REPAY_STS_MAY 593.383490 7 REPAY_STS_APR 506.173542 0 LIMIT_BAL 76.341829 24 EDUCATION_Others 61.686112 22 EDUCATION_Graduate School 51.149551 21 SEX_Male 28.922107 23 EDUCATION_High School 26.336315 25 EDUCATION_University 21.237973 20 SEX_Female 18.983326 26 MARRIAGE_Married 14.486745 28 MARRIAGE_Single 13.159406 18 PAY_AMT_MAY 10.394332 19 PAY_AMT_APR 9.725684 14 PAY_AMT_SEPT 8.847610 17 PAY_AMT_JUN 7.933442 16 PAY_AMT_JUL 6.284283 15 PAY_AMT_AUG 5.479844 1 AGE 0.585334 27 MARRIAGE_Others 0.484174 8 BILL_AMT_SEPT 0.256197 9 BILL_AMT_AUG 0.244231 10 BILL_AMT_JUL 0.076826 11 BILL_AMT_JUN 0.056570 12 BILL_AMT_MAY 0.041310 13 BILL_AMT_APR 0.006235
import pandas as pd
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import mutual_info_classif
from sklearn.preprocessing import OneHotEncoder
# Load your dataset (replace 'credit.csv' with the actual path to your dataset)
data = df
# Assuming your target variable is 'IsDefaulter' and other columns are features
X = data.drop('IsDefaulter', axis=1)
y = data['IsDefaulter']
# One-hot encode categorical variables
X = pd.get_dummies(X)
# Initialize the SelectKBest object with mutual_info_classif as the scoring function
k_best = SelectKBest(score_func=mutual_info_classif, k='all')
# Fit the SelectKBest object to your data
k_best.fit(X, y)
# Get the feature scores
feature_scores = k_best.scores_
# Create a DataFrame to display the feature names and their corresponding information gain scores
feature_info_gain = pd.DataFrame({'Feature': X.columns, 'Information_Gain': feature_scores })
# Sort the features by information gain in descending order
feature_info_gain = feature_info_gain.sort_values(by='Information_Gain', ascending=False)
#Display the top features with their information gain scores
print(feature_info_gain)
Feature Information_Gain 2 REPAY_STS_SEPT 0.075546 3 REPAY_STS_AUG 0.042527 4 REPAY_STS_JUL 0.035761 5 REPAY_STS_JUN 0.030931 6 REPAY_STS_MAY 0.030359 7 REPAY_STS_APR 0.027707 14 PAY_AMT_SEPT 0.020531 16 PAY_AMT_JUL 0.018094 17 PAY_AMT_JUN 0.017145 0 LIMIT_BAL 0.016788 15 PAY_AMT_AUG 0.015735 18 PAY_AMT_MAY 0.013298 8 BILL_AMT_SEPT 0.011355 19 PAY_AMT_APR 0.010555 10 BILL_AMT_JUL 0.008844 13 BILL_AMT_APR 0.007274 12 BILL_AMT_MAY 0.006047 9 BILL_AMT_AUG 0.004605 11 BILL_AMT_JUN 0.004046 22 EDUCATION_Graduate School 0.003778 24 EDUCATION_Others 0.003767 20 SEX_Female 0.002233 26 MARRIAGE_Married 0.001159 25 EDUCATION_University 0.000967 23 EDUCATION_High School 0.000520 1 AGE 0.000000 21 SEX_Male 0.000000 27 MARRIAGE_Others 0.000000 28 MARRIAGE_Single 0.000000
Do you think that your data need to be transformed? If yes, which transformation have you used. Explain why?
# Extracting independent and dependent variables
X = data.drop(labels='IsDefaulter', axis=1)
Y = data['IsDefaulter']
# One-hot encoding categorical variables
X = pd.get_dummies(X)
# Print the shape of X and Y
print(f"The number of rows and columns in X is {X.shape} respectively.")
print(f"The number of rows in Y is {Y.shape}.")
The number of rows and columns in X is (30000, 29) respectively. The number of rows in Y is (30000,).
# Scaling your data
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
# Now you can use X_scaled for further processing such as modeling
X_scaled
array([[-1.13672015, -1.24601985, 2.16049607, ..., 1.09377971,
-0.11281222, -1.06647132],
[-0.3659805 , -1.02904717, -0.46907093, ..., -0.91426088,
-0.11281222, 0.93767172],
[-0.59720239, -0.16115646, -0.46907093, ..., -0.91426088,
-0.11281222, 0.93767172],
...,
[-1.05964618, 0.16430256, 4.79006307, ..., -0.91426088,
-0.11281222, 0.93767172],
[-0.67427636, 0.59824792, 0.84571257, ..., 1.09377971,
-0.11281222, -1.06647132],
[-0.90549825, 1.14067961, -0.46907093, ..., 1.09377971,
-0.11281222, -1.06647132]])
Which method have you used to scale you data and why?
I've used standard scaler to the data so that its magnitudes are normalized. This could help the learning algorithm and speed up calculations
#Split your data to train and test. Choose splitting ration wisely
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size = 0.2, random_state = 42)
What data splitting ratio have you used and why?
I have made the use of 80% and 20% Training and Testing data respectively. The data should be divided in such a way that neither of them is too high, which is more dependent on the ammount of data we have. If the data is too small, then no split will give satisfactory variance so we'll have to do cross-validation, but if the data is huge then it doesn't really matter whether we choose an 80:20 split or a 90:10 split (indeed we may choose to use less training data as otherwise, it might be more computationally intensive).
credit_df['IsDefaulter'].value_counts()
No 23364 Yes 6636 Name: IsDefaulter, dtype: int64
Do you think dataset is imbalaced? Explain why
Yes, definitely there is class imbalance in the Target variable. However we have already visualized the same in the exploratory data analysis section by using variour charts
# Handling Imbalanced Dataset (If needed)
# Let's again check the distribution of the target variable
plt.figure(figsize=(10,5))
ax = sns.countplot(data=credit_df, x=credit_df['IsDefaulter'], palette='Set3')
ax.set_title('IsDefaulter vs Non-Defaulter', fontsize=20)
ax.set_title('Status', fontsize=15)
ax.set_ylabel('count', fontsize=15)
ax.set_xticklabels(['Yes', 'No'], fontsize =13)
# adding percentage to the top of each bar
total = float(len(credit_df))
for p in ax.patches:
height = p.get_height()
ax.text(p.get_x() + p.get_width()/2.,
height + 3,
'{:1.2f}%'.format(height/total * 100),
ha="center")
What technique did you use to handle the imbalance dataset and why?
As we can see that there is a class imbalance in the dataset so we will tackle the class imbalance using machine learning algorithm
ML Model - 1
from sklearn import metrics
# ML model - 1 Implementation
knn = KNeighborsClassifier(n_neighbors=6, weights='distance')
#Fit the Algorithm
knn.fit(X_train, Y_train)
# Predict on the model
knn_y_pred_train = knn.predict(X_train)
knn_tree_y_pred_test = knn.predict(X_test)
print('Untuned K Nearest Neighbors Model')
print("Training F1 Score: ", metrics.f1_score(Y_train, knn_y_pred_train, pos_label='Yes'))
print("Testing F1 Score: ", metrics.f1_score(Y_test, knn_tree_y_pred_test, pos_label='Yes'))
Untuned K Nearest Neighbors Model Training F1 Score: 0.9987773911407881 Testing F1 Score: 0.2494908350305499
t1_knn = metrics.f1_score(Y_train, knn_y_pred_train, pos_label='Yes')
t2_knn = metrics.f1_score(Y_test, knn_tree_y_pred_test, pos_label='Yes')
#Visualizing evaluation Metric Score Chart
metric_dict = {}
metric_dict['without cross KNN'] = {'Training F1':t1_knn, 'Testing F1': t2_knn}
from sklearn.model_selection import GridSearchCV
#ML model - Implementation with hyperparameter optimization techniques (i.e GridSearch CV, Randomsearch CV, Bayesian )
k_near = KNeighborsClassifier(weights='distance')
#Fit the Algorithm
k_param_dict = {'n_neighbors': range(4, 15, 1)}
grid_k_neighbors = GridSearchCV(k_near, k_param_dict, cv=5, scoring='f1', verbose=1)
grid_k_neighbors.fit(X_train, Y_train)
#Predict on the model
print('F1 Score:', grid_k_neighbors.best_score_)
print('Best Herperparameters:', grid_k_neighbors.best_params_)
print('Model object with best parameters: ')
print(grid_k_neighbors.best_estimator_)
Fitting 5 folds for each of 11 candidates, totalling 55 fits
F1 Score: nan
Best Herperparameters: {'n_neighbors': 4}
Model object with best parameters:
KNeighborsClassifier(n_neighbors=4, weights='distance')
#Predicting the response on both train and test data set respectively
tuned_k_neighbors_y_pred_train = grid_k_neighbors.best_estimator_.predict(X_train)
tuned_k_neighbors_y_pred_test = grid_k_neighbors.best_estimator_.predict(X_test)
# Calculate F1 score for train and test sets
print('Tuned K Nearest Neighbors Predictions')
print('F1 on train set:', metrics.f1_score(Y_train, tuned_k_neighbors_y_pred_train, pos_label='Yes'))
print('F1 on test set:', metrics.f1_score(Y_test, tuned_k_neighbors_y_pred_test, pos_label='Yes'))
Tuned K Nearest Neighbors Predictions F1 on train set: 0.9987773911407881 F1 on test set: 0.27683873915029694
train_tuned_knn = metrics.f1_score(Y_train, tuned_k_neighbors_y_pred_train, pos_label='Yes')
test_tuned_knn = metrics.f1_score(Y_test, tuned_k_neighbors_y_pred_test, pos_label='Yes')
metric_dict = {}
metric_dict['Tuned KNN'] = {'Training F1':train_tuned_knn, 'Testing F1':test_tuned_knn}
metric_dict['Tuned KNN']
{'Training F1': 0.9987773911407881, 'Testing F1': 0.27683873915029694}
Which Hyperparameter optimization technique have you used and why?
Since hyperparameter values are predefined so here grid search technique has been used to tune the hyper parameter.
Have you seen any improvement? Note down the improvement with updates Evaluation metric Score Chart
Yes, there is a slight improvement in the F1 score evaluation matrics. But it is very much clear that the above model is definitely not suitable for the predicting the default as the high evaluation metric F1 score in the train data indicates that the model has overfit in the data set.
ML Model - 2
Explain the ML Model used and it's performance using Evaluation metric Score chart.
logreg = LogisticRegression(C=1e9, class_weight='balanced')
logreg.fit(X_train, Y_train)
logreg_y_pred_train = logreg.predict(X_train)
logreg_y_pred_test = logreg.predict(X_test)
print('Untuned Logistic Regression Model')
print('Training F1 Score: ', metrics.f1_score(Y_train, logreg_y_pred_train, pos_label='Yes'))
print('Testing F1 Score: ', metrics.f1_score(Y_test, logreg_y_pred_test, pos_label='Yes'))
Untuned Logistic Regression Model Training F1 Score: 0.398454825094785 Testing F1 Score: 0.38547326279668287
t1_log_reg = metrics.f1_score(Y_train, logreg_y_pred_train, pos_label='Yes')
t2_log_reg = metrics.f1_score(Y_test, logreg_y_pred_test, pos_label='Yes')
#Visualizing Evaluation Metric Score Chart
metric_dict['without cross Logistic Regression'] = {'Training F1': t1_log_reg, 'Testing F1':t2_log_reg}
y_pred_prob = logreg.predict_proba(X_test)[:, 1]
# ROC Curve
fpr, tpr, thresholds = metrics.roc_curve(Y_test, y_pred_prob, pos_label='Yes')
# Plot ROC Curve
plt.plot(fpr, tpr)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.title('ROC Curve for K Nearest Neighbors')
plt.xlabel('False Positive Rate (1 - Specificity)')
plt.ylabel('True Positive Rate (Sensitivity)')
plt.grid(True)
plt.show()
print(metrics.roc_auc_score(Y_test, y_pred_prob))
0.6385903808414355
print("Accuracy:", metrics.accuracy_score(Y_test, logreg_y_pred_test))
print('Precision:', metrics.precision_score(Y_test, logreg_y_pred_test, pos_label='Yes'))
print('Recall:', metrics.recall_score(Y_test, logreg_y_pred_test, pos_label='Yes'))
print('F1 Score:', metrics.f1_score(Y_test, logreg_y_pred_test, pos_label='Yes'))
Accuracy: 0.6418333333333334 Precision: 0.3086080586080586 Recall: 0.5133282559025133 F1 Score: 0.38547326279668287
#confuse matrix
from sklearn.metrics import confusion_matrix
labels = ['Not default', 'default']
cm = confusion_matrix(Y_train, logreg_y_pred_train)
print(cm)
ax = plt.subplot()
sns.heatmap(cm, annot=True, ax=ax) #annot=True to annotate cells
#labels, title and ticks
ax.set_xlabel('Predicted labels')
ax.set_ylabel('True labels')
ax.set_title('Confusion Matrix')
ax.xaxis.set_ticklabels(labels)
ax.yaxis.set_ticklabels(labels)
[[12806 5871] [ 2538 2785]]
[Text(0, 0.5, 'Not default'), Text(0, 1.5, 'default')]
labels = ['Not default', 'Default']
cm = confusion_matrix(Y_test, logreg_y_pred_test)
print(cm)
ax = plt.subplot()
sns.heatmap(cm, annot=True, ax=ax) #annot=True to annotate cells
#labels, title and ticks
ax.set_xlabel('Predicted labels')
ax.set_ylabel('True labels')
ax.set_title('Confusion Matrix')
ax.xaxis.set_ticklabels(labels)
ax.yaxis.set_ticklabels(labels)
[[3177 1510] [ 639 674]]
[Text(0, 0.5, 'Not default'), Text(0, 1.5, 'Default')]
Cross Validation & Hyperparameter Tuning
#ML Model -1 Implementation with hyperparameter optimization techniques (i.e GridSearch CV, Randomsearch CV, Bayesian Optimization etc.)
tuned_log = LogisticRegression(random_state=42, class_weight='balanced', solver='lbfgs')
# Fit the Algorithm
log_param_dict = {'C': range(3, 7, 1), 'penalty': ['l2']}
grid_log = GridSearchCV(tuned_log, log_param_dict, cv=5, scoring='f1', verbose=1)
grid_log.fit(X_train, Y_train)
Fitting 5 folds for each of 4 candidates, totalling 20 fits
GridSearchCV(cv=5,
estimator=LogisticRegression(class_weight='balanced',
random_state=42),
param_grid={'C': range(3, 7), 'penalty': ['l2']}, scoring='f1',
verbose=1)
print('F1 Score: ', grid_log.best_score_)
print('Best Hyperparameters:', grid_log.best_params_)
print('Model object with best parameters: ')
print(grid_log.best_estimator_)
F1 Score: nan
Best Hyperparameters: {'C': 3, 'penalty': 'l2'}
Model object with best parameters:
LogisticRegression(C=3, class_weight='balanced', random_state=42)
grid_log_y_pred_train = grid_log.best_estimator_.predict(X_train)
grid_log_y_pred_test = grid_log.best_estimator_.predict(X_test)
print("Tuned Logistic Regression Model Prediction")
print("F1 on train set: ", metrics.f1_score(Y_train, grid_log_y_pred_train, pos_label='Yes'))
print("F1 on test set: ", metrics.f1_score(Y_test, grid_log_y_pred_test, pos_label='Yes'))
Tuned Logistic Regression Model Prediction F1 on train set: 0.393684521174561 F1 on test set: 0.3854566952409104
train_tuned_log_reg = metrics.f1_score(Y_train, grid_log_y_pred_train, pos_label='Yes')
test_tuned_log_reg = metrics.f1_score(Y_test, grid_log_y_pred_test, pos_label='Yes')
metric_dict['Tuned Logistic Regression'] = {'Training F1': train_tuned_log_reg, 'Testing F1':test_tuned_log_reg}
metric_dict['Tuned Logistic Regression']
{'Training F1': 0.393684521174561, 'Testing F1': 0.3854566952409104}
labels = ['Not deault', 'default']
cm = confusion_matrix(Y_test, grid_log_y_pred_test)
print(cm)
ax = plt.subplot()
sns.heatmap(cm, annot=True, ax = ax) #annot = True to annotate cells
#label, title and ticks
ax.set_xlabel('Predicted labels')
ax.set_ylabel('True labels')
ax.set_title('Confusion Matrix')
ax.xaxis.set_ticklabels(labels)
[[3269 1418] [ 661 652]]
[Text(0.5, 0, 'Not deault'), Text(1.5, 0, 'default')]
y_pred_prob_1 = grid_log.predict_proba(X_test)[:, 1]
#ROC Curve Graph
fpr, tpr, thresholds = metrics.roc_curve(Y_test, y_pred_prob_1, pos_label='Yes')
plt.plot(fpr, tpr)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.title('ROC curve for Logistic Regression')
plt.xlabel('False Positive Rate (1 - Specificity)')
plt.ylabel('True Positive Rate (Sensitivity)')
plt.grid(True)
print(metrics.roc_auc_score(Y_test, y_pred_prob_1))
0.6405934581740002
Which hyperparameter optimization technique have you used and why?
Grid search CV technique has been issued for hyperparameter tuning.
Have you seen any improvement? Note down the improvement with updates Evaluation metric Score Chart
There is not much improvement noticed even after applying grid search CV. F1 score and roc_suc score remains same for both the cases.
ML - Model 3
tree = DecisionTreeClassifier()
tree.fit(X_train, Y_train)
DecisionTreeClassifier()
#predicting the model
tree_y_pred_train = tree.predict(X_train)
tree_y_pred_test = tree.predict(X_test)
print('Untuned Decision Tree Model')
print('Training F1 Score: ', metrics.f1_score(Y_train, tree_y_pred_train, pos_label='Yes'))
print('Testing F1 Score: ', metrics.f1_score(Y_test, tree_y_pred_test,pos_label='Yes'))
Untuned Decision Tree Model Training F1 Score: 0.9987773911407881 Testing F1 Score: 0.39163916391639164
t1_tree = metrics.f1_score(Y_train, tree_y_pred_train, pos_label='Yes')
t2_tree = metrics.f1_score(Y_test, tree_y_pred_test, pos_label='Yes')
metric_dict['without corss Decision Tree'] = {'Training F1':t1_tree, 'Testing F1': t2_tree}
#ML model - 3 Implementation with hyperparameter optimization techniques(i.e GridSearch CV, Randomsearch CV, Bayesian Optimization etc.)
decision_tree = DecisionTreeClassifier(random_state=3, criterion='gini')
param_dict = {'max_depth': range(2, 15, 1),
'max_leaf_nodes': range(2, 30, 2),
'min_samples_split': range(2, 4, 1)}
grid_tree = GridSearchCV(decision_tree, param_dict, cv=5, scoring='f1', verbose=1)
#Fit the Algorithm
grid_tree.fit(X_train, Y_train)
Fitting 5 folds for each of 364 candidates, totalling 1820 fits
GridSearchCV(cv=5, estimator=DecisionTreeClassifier(random_state=3),
param_grid={'max_depth': range(2, 15),
'max_leaf_nodes': range(2, 30, 2),
'min_samples_split': range(2, 4)},
scoring='f1', verbose=1)
print('F1 Score:', grid_tree.best_score_)
print('Best Hyperparameters:', grid_tree.best_params_)
print('Model object with the best parameters: ')
print(grid_tree.best_estimator_)
F1 Score: nan
Best Hyperparameters: {'max_depth': 2, 'max_leaf_nodes': 2, 'min_samples_split': 2}
Model object with the best parameters:
DecisionTreeClassifier(max_depth=2, max_leaf_nodes=2, random_state=3)
#predict on the model
tuned_tree_y_pred_train = grid_tree.best_estimator_.predict(X_train)
tuned_tree_y_pred_test = grid_tree.best_estimator_.predict(X_test)
print('Tuned Decision Tree Model Prediction')
print('F1 on train set: ', metrics.f1_score(Y_train, tuned_tree_y_pred_train, pos_label='Yes'))
print('F1 on test set: ', metrics.f1_score(Y_test, tuned_tree_y_pred_test,pos_label='Yes'))
Tuned Decision Tree Model Prediction F1 on train set: 0.44838134081060416 F1 on test set: 0.4354166666666667
train_tuned_tree = metrics.f1_score(Y_train, tuned_tree_y_pred_train, pos_label='Yes')
test_tuned_tree = metrics.f1_score(Y_test, tuned_tree_y_pred_test, pos_label='Yes')
metric_dict['Tuned Decision Tree'] = {'Training F1': train_tuned_tree, 'Testing F1': test_tuned_tree}
Which hyperparameter optimization techniques have you used and why?
Since hyperparameter values are predefined so here grid search techniques has been used to tune the hyperparameter
Have you seen any improvement? Note down the improvement with updates Evaluation metric Score Chart
Yes there is an improvement after using cross validation. without cross validation the model was overfit that's the reason the evaluation matrics was quite high in the train data but after using cross validation the matrics numbers were balanced for both train and test data
Model - 4
#Applying SMOTE technique to tackle the class imbalance
from imblearn.over_sampling import SMOTE
smote = SMOTE(random_state = 42)
X_res, Y_res = smote.fit_resample(X_train, Y_train)
print('Original dataset shape', len(credit_df))
print('Resampled dataset shape', len(Y_res))
Original dataset shape 30000 Resampled dataset shape 37354
import xgboost
from sklearn.model_selection import RandomizedSearchCV
model = xgboost.XGBClassifier()
params = {
'learning_rate': [0.05, 0.10, 0.15, 0.20, 0.25, 0.30],
'max_depth': [3, 4, 5, 6, 8, 10, 12, 15],
'min_child_weight': [1, 3, 5, 7],
'gamma': [0.0, 0.1, 0.2, 0.3, 0.4],
'colsample_bytree': [0.3, 0.4, 0.5, 0.7]
}
random_search = RandomizedSearchCV(model, param_distributions = params, n_iter = 5, scoring='f1', n_jobs=-1, cv=5, verbose=3)
Y_res_numeric = Y_res.replace({'Yes': 1, 'No': 0})
random_search.fit(X_res, Y_res_numeric)
Fitting 5 folds for each of 5 candidates, totalling 25 fits
RandomizedSearchCV(cv=5,
estimator=XGBClassifier(base_score=None, booster=None,
callbacks=None,
colsample_bylevel=None,
colsample_bynode=None,
colsample_bytree=None, device=None,
early_stopping_rounds=None,
enable_categorical=False,
eval_metric=None, feature_types=None,
gamma=None, grow_policy=None,
importance_type=None,
interaction_constraints=None,
learning_rate...
monotone_constraints=None,
multi_strategy=None,
n_estimators=None, n_jobs=None,
num_parallel_tree=None,
random_state=None, ...),
n_iter=5, n_jobs=-1,
param_distributions={'colsample_bytree': [0.3, 0.4, 0.5,
0.7],
'gamma': [0.0, 0.1, 0.2, 0.3, 0.4],
'learning_rate': [0.05, 0.1, 0.15, 0.2,
0.25, 0.3],
'max_depth': [3, 4, 5, 6, 8, 10, 12,
15],
'min_child_weight': [1, 3, 5, 7]},
scoring='f1', verbose=3)
#Best Estimators:
random_search.best_estimator_
XGBClassifier(base_score=None, booster=None, callbacks=None,
colsample_bylevel=None, colsample_bynode=None,
colsample_bytree=0.5, device=None, early_stopping_rounds=None,
enable_categorical=False, eval_metric=None, feature_types=None,
gamma=0.3, grow_policy=None, importance_type=None,
interaction_constraints=None, learning_rate=0.1, max_bin=None,
max_cat_threshold=None, max_cat_to_onehot=None,
max_delta_step=None, max_depth=10, max_leaves=None,
min_child_weight=1, missing=nan, monotone_constraints=None,
multi_strategy=None, n_estimators=None, n_jobs=None,
num_parallel_tree=None, random_state=None, ...)
#best_param
random_search.best_params_
{'min_child_weight': 1,
'max_depth': 10,
'learning_rate': 0.1,
'gamma': 0.3,
'colsample_bytree': 0.5}
model_ = xgboost.XGBClassifier(obejective='binary:logistic',
min_child_weight=3,
max_depth=10,
learning_rate=0.05,
gamma=0.4,
colsample_bynode=1,
colsample_bytree=0.7,
use_label_encoder=False)
Y_res_numeric = Y_res.replace({'Yes': 1, 'No': 0})
#Fitting the model
model_.fit(X_res, Y_res_numeric)
XGBClassifier(base_score=None, booster=None, callbacks=None,
colsample_bylevel=None, colsample_bynode=1, colsample_bytree=0.7,
device=None, early_stopping_rounds=None, enable_categorical=False,
eval_metric=None, feature_types=None, gamma=0.4, grow_policy=None,
importance_type=None, interaction_constraints=None,
learning_rate=0.05, max_bin=None, max_cat_threshold=None,
max_cat_to_onehot=None, max_delta_step=None, max_depth=10,
max_leaves=None, min_child_weight=3, missing=nan,
monotone_constraints=None, multi_strategy=None, n_estimators=None,
n_jobs=None, num_parallel_tree=None, obejective='binary:logistic', ...)
# Predicting using the trained model_
x_pred_train = model_.predict(X_res) # Predictions on training data
x_pred = model_.predict(X_test) # Predictions on test data
# Convert true labels to numeric values
Y_res_numeric = Y_res.replace({'Yes': 1, 'No': 0})
Y_test_numeric = Y_test.replace({'Yes': 1, 'No': 0})
# Compute F1 score using numeric labels
f1_train = metrics.f1_score(Y_res_numeric, x_pred_train)
f1_test = metrics.f1_score(Y_test_numeric, x_pred)
print("F1 on train set:", f1_train)
print("F1 on test set:", f1_test)
F1 on train set: 0.91261860121584 F1 on test set: 0.4906520747834018
# Compute F1 score using numeric labels
train_tuned_XGB = metrics.f1_score(Y_res_numeric, x_pred_train)
test_tuned_XGB = metrics.f1_score(Y_test_numeric, x_pred)
metric_dict['Tuned XGB'] = {'Training F1': train_tuned_XGB, 'Testing F1': test_tuned_XGB}
# Compute confusion matrix using numeric labels
cm = confusion_matrix(Y_test_numeric, x_pred)
print(cm)
# Plot confusion matrix
labels = ['Not default', 'Default']
ax = plt.subplot()
sns.heatmap(cm, annot=True, ax=ax) # annot = True to annotate cells
# Labels, title, and ticks
ax.set_xlabel("Predicted labels")
ax.set_ylabel('True labels')
ax.set_title('Confusion Matrix')
ax.xaxis.set_ticklabels(labels)
ax.yaxis.set_ticklabels(labels)
[[4345 342] [ 775 538]]
[Text(0, 0.5, 'Not default'), Text(0, 1.5, 'Default')]
y_pred_prob_ = random_search.predict_proba(X_test)[:, 1]
#ROC Curve graph
fpr, tpr, thresholds = metrics.roc_curve(Y_test, y_pred_prob_, pos_label='Yes')
plt.plot(fpr, tpr)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.title("ROC curve for XGB classifier")
plt.xlabel('False Positive Rate (1 - Specificity)')
plt.ylabel("True Positive Rate (Sensitivity)")
plt.grid(True)
print(metrics.roc_auc_score(Y_test, y_pred_prob_))
0.7617483564837422
Model - 5
from sklearn.ensemble import RandomForestClassifier
Rfc = RandomForestClassifier(random_state=5, criterion='gini', class_weight='balanced')
Rfc_param_dict = {'n_estimators':[50, 100],
'max_depth': list(range(2, 15)),
'max_features': list(range(5, 15))}
grid_forest = GridSearchCV(Rfc, Rfc_param_dict, cv=5, scoring='f1', verbose=1, n_jobs=2)
grid_forest.fit(X_train, Y_train)
Fitting 5 folds for each of 260 candidates, totalling 1300 fits
GridSearchCV(cv=5,
estimator=RandomForestClassifier(class_weight='balanced',
random_state=5),
n_jobs=2,
param_grid={'max_depth': [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13,
14],
'max_features': [5, 6, 7, 8, 9, 10, 11, 12, 13, 14],
'n_estimators': [50, 100]},
scoring='f1', verbose=1)
print(grid_forest.best_score_)
print(grid_forest.best_params_)
print(grid_forest.best_estimator_)
nan
{'max_depth': 2, 'max_features': 5, 'n_estimators': 50}
RandomForestClassifier(class_weight='balanced', max_depth=2, max_features=5,
n_estimators=50, random_state=5)
# Predict the testing set
forest_y_pred_train = grid_forest.best_estimator_.predict(X_train)
# Predict the training set
forest_y_pred_test = grid_forest.best_estimator_.predict(X_test)
print("F1 on train set: ", metrics.f1_score(Y_train, forest_y_pred_train, pos_label='Yes'))
print("F1 on test set: ", metrics.f1_score(Y_test, forest_y_pred_test, pos_label='Yes'))
F1 on train set: 0.5291828793774319 F1 on test set: 0.5284580889962057
train_tuned_forest = metrics.f1_score(Y_train, forest_y_pred_train, pos_label='Yes')
test_tuned_forest = metrics.f1_score(Y_test, forest_y_pred_test, pos_label='Yes')
metric_dict['Tuned Random Forest'] = {'Training F1': train_tuned_forest, 'Testing F1': test_tuned_forest}
labels = ['Not default', 'default']
cm = confusion_matrix(Y_test, forest_y_pred_test)
print(cm)
ax = plt.subplot()
sns.heatmap(cm, annot=True, ax = ax) #annot = True to annotate cells
#labels, title and ticks
ax.set_xlabel('Predicted labels')
ax.set_ylabel('True labels')
ax.set_title('Confusion Matrix')
ax.xaxis.set_ticklabels(labels)
ax.yaxis.set_ticklabels(labels)
[[3867 820] [ 547 766]]
[Text(0, 0.5, 'Not default'), Text(0, 1.5, 'default')]
y_pred_prob_2 = grid_forest.predict_proba(X_test)[:, 1]
y_pred_prob_2
array([0.4103467 , 0.36421357, 0.38792973, ..., 0.42863211, 0.54500857,
0.41331008])
#ROC Curve graph
fpr, tpr, threshold = metrics.roc_curve(Y_test, y_pred_prob_2, pos_label='Yes')
plt.plot(fpr, tpr)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.title('ROC curve for Randomforest classifier')
plt.xlabel('False Positive Rate (1 - Specificity)')
plt.ylabel('True Positive Rate (Sensitivity)')
plt.grid(True)
print(metrics.roc_auc_score(Y_test, y_pred_prob_2))
0.7592534064258045
Evaluation_metric_comparison = pd.DataFrame.from_dict(metric_dict, orient='index')
Evaluation_metric_comparison
| Training F1 | Testing F1 | |
|---|---|---|
| Tuned KNN | 0.998777 | 0.276839 |
| without cross Logistic Regression | 0.398455 | 0.385473 |
| Tuned Logistic Regression | 0.393685 | 0.385457 |
| without corss Decision Tree | 0.998777 | 0.391639 |
| Tuned Decision Tree | 0.448381 | 0.435417 |
| Tuned XGB | 0.912619 | 0.490652 |
| Tuned Random Forest | 0.529183 | 0.528458 |
Since there is a high class imbalance so I have choosen F1 score as evaluation metrics for the positive business impact. The F1 score is concerened about precision and recall both
As per the above implementation of modes the Random forest model has the best F1 score as the other models is not performing good so I would choose Random Forest Model.
There was a high class imbalance in target variable
I have used weights distance, balanced and SMOTE techniques to tackle the class imbalance
I have implemented 5 classification models and chieved a fairly good results for only one algorithm due to highly class imbalance issue